Abstract Twenty-four albino rats were tested in a two-lever situation in which one lever was associated with a condition of discontinuous negatively correlated reinforcement (reward for latency of 3.5 sec or longer), while response to the other lever was reinforced, regardless of latency, with a yoked frequency equal to that determined by the latencies on correlated-lever trials. Sixty-six acquisition, 15 extinction, and 10 reacquisition sessions were given, with 50% random reinforcement on each lever during reacquisition. In acquisition most of the rats showed eventual preference for the yoked-uncorrelated lever in terms of choices on two-lever trials and speeds on one-lever forced trials. An early preference for the correlated lever was displayed by over half of the rats and maintained over most of the acquisition period by the slowest-learning rats. Relative rate of extinction was essentially equal for the two levers. Historical effects of the acquisition conditions appeared with remarkable persistence in reacquisition.